DComboNet is an R package for personalized anti-cancer drug combination prediction based on multi-level data integration. There are two main prediction models contained in the package. The level one model is for generalized anti-cancer drug combination effectiveness prediction and level two model is for cancer sample specific drug combination prediction. The two-level model based on a network-based method which integrates five subnetwork including drug-drug, drug-gene, drug-pathway, gene-gene and pathway-pathway association networks. Random walk with restart(RWR) algorithm is used to capture global proximity between drugs and the result is based on the rank returned via RWR algorithm. DComboNet also provide clues for the potential mechanisms of drug combinations by extracting the top ranked genes/drugs between predicted drug combinations.
This tutorial provides the instruction of the main usage of this package fitting different scenario. This tutorial will lead to know the basic usage of the prediction, the description of prepared data how to extend the network construction to fit your own dataset. This tutorial will not present detail description of all functions contain in the packagem but you can easily learn those in help document with the R package.
DComboNet has been upload in Github and can be install as follow:
install.packages("devtools")
devtools::install_github(".../DComboNet")
library(DComboNet)The prediction of drug combinable tendency is packed in function DComboNet. It provides two level prediction models and includes network construction, transition matrix generating and global similarity calculation:
DComboNet(load_dir,
resultdir,
model = c('L1','L2'),
drugcandidate = NULL,
manual_input = c('TRUE','T','FALSE','F'),
CDK_FP = NULL,
pubchemFP = NULL,
MACCS_FP = NULL,
drugnetWeight = c('TRUE','T','FALSE','F'),
featuretype = c('ATCsim','STsim','SEsim','integrated_score'),
drugtarget = NULL,
druggene = NULL,
dataset = NULL,
cellline = NULL,
treatment_time = NULL,
foldchange_DEG = NULL,
pvalue_DEG = NULL,
foldchange_DEP = NULL,
pvalue_DEP = NULL,
drugDEG = NULL,
drugDEP = NULL,
cancergene = NULL,
dtweight = 2,
dgweight = 1,
dDEGweight = 1,
dgpweight = 1,
dDEpweight = 1,
gpweight=1,
x = 0.5,
y = 0.5,
z = 0,
A = 0.5,
B = 0.5,
eta = 1,
r = 0.7)\(\bullet\) The function provide internal datasets for network construction. User should provide the path (load_dir) to access the data folder downloaded with the package.
\(\bullet\) Path to save result files should be specified in result_dir.
\(\bullet\) Two level models are provided for combination prediction, L1 and L2 denote level one and level two model, respectively. Different models require differnent following inputs. The details about differnet usages of the two-level models will describes in next session.
\(\bullet\) Two solutions are provided here for users to input their interested drugs. If manual_input = TRUE, after seeing the tip “Please type in the drug you are interested:”, user can manually input drug name; if manual_input = FALSE, drugcandidate should be feed with data.frame contain drug name(s) and their drugbank ID. Note that if the drug you are interested in are not contained in the provided drug list, drugcandidate,drugpair (drugpair that contains newly added druglist) together with their chemical fingerprints generated via PADEL (CDK fingerprint CDK_FP, pubchem fingerprint pubchemFP and MACCS fingerprint MACCS_FP) should be provided for adding newly inputed drug(s) in the drug-drug association network.
\(\bullet\) Next two parameters drugnetWeight and featuretype are necessary. User needs to inform whether it is necessary to assign values to the edge of drug-drug association network and which feature should be used to assign. Three drug-drug similarities(ATCsim, STsim, SEsim) together with integrated pharmacological score (integrated_score) are provided as options.
\(\bullet\) To update drug-gene association network, gene-gene association network and drug-pathway assocition netwrok, if new drug(s) was taken as input in drugcandidate, User must provide drug(s) and their target genes table in drugtarget for both level one and level two models. For level one model, druggene can be provided to extend the connections between drug and gene or pathways. Here, druggene table can be extracted from multiple resources, such as IPA tools and publications. For level two model, cancer sample specific expressed gene list and drug induced transcriptome change should be provided. DComboNet() gives two options: if the cell line and drug(s) are in our provided datasets (which is generated from LINCS database and CCLE database), user can put in cellline, dataset and treatment_time to create network. Differentially expressed genes were selected by functions lmFIt() and eBayes() in the Limma package. Differential regulated pathways (DEpathway) were obtained by the GSVA algorithm. Note that gene expression filtering criteria (foldchange and pvalue) provides a detailed customizing solution. If cell line or drugs are not in provided dataset, you can simply provided your own cancer sample/cell line specific expressed gene list and drug-induced differentially expressed gene list in cancergene and drugDEG.
\(\bullet\) Other then weight method for drug-drug assocaition network described above, for drug-gene association, drug-pathway association and gene-pathway association, different edge weight strategies are made as parameters. For level one model, the default weights are:
\(\cdot\) dtweight = 2: the edge weight of drug-target gene associations sets as 2.
\(\cdot\) dgweight = 1: the edge weight of drug-related gene associations sets as 1.
\(\cdot\) dgpweight = 1: the edge weight of drug-pathway associations sets as 1.
For level two model, drug-targeted gene association will be weighted with the value set to dtweight, and drug-DEG association will be weighted according to the foldchange between drug-induced gene expression profile and control; the associations between drugs and their differentially regulated pathways will be weighted by parameter dDEpWeight.
\(\bullet\) Parameters x, y, z, A and B denote the jumping probability in different subnetworks) and these jumping probabilities are not independent (\(x=1-A\), \(y=1-A\), \(B=1-x\)):
\(\cdot\) x denotes the jumping probability in drug-gene association network (inter-network crossing event happens between drug and gene nodes).
\(\cdot\) y denotes the jumping probability in drug-pathway association network (inter-network crossing event happens between gene and pathway nodes).
\(\cdot\) z denotes the jumping probability in gene-pathway association network.
\(\cdot\) A denotes the jumping probability in drug-drug association network (when the inter-network crossing events between drug and gene nodes and between drug and pathway nodes are neither existing).
\(\cdot\) B denotes the jumping probability in gene-gene assocaition network (when the inter-network crossing events between drug and gene nodes and between gene and pathway nodes are neither existing).
\(\bullet\) Numeric parameter eta is to controls the probability of restarting in the corresponding network. in DComboNet(), it always starts from drug seed in drug-drug association network, therefore eta sets to be 1.
\(\bullet\) Numeric parameter r denotes global restart probability. It can be set as any value from 0 to 1, but according to previous researches and our tuning result, 0.7 shows the best prediction power.
To quickly get the usage of DComboNet, prediction assignment for one drug is taken as an example to go through the basic functions. Here, Sorafenib is taken as example, note that Sorafenib is within our pre-prepared data set, if you want to know how to do prediction for new drugs, please check session 5 for more details.
Before start any prediction assignment, please make sure data folder is fully downloaded and put in a fetchable path. data folder contains data files for construct networks as well as testing datasets used in our publication.
# 1. Make sure the path of `data` folder assign to variable correctly
## Here is an example path, you should switch to your own
load_dir = "G:/lab/DCcomboNet/Rpackage/input_data/"
# 2. Specify the path to save result files
## Here is an example path, you should switch to your own
resultdir = "G:/lab/DCcomboNet/Rpackage/tryout_result/"
As we introduced in session 3, DComboNet() function incorporate two level models, you can choose level one model which is a generalized anti-cancer drug combination model and level two model which is a cancer-sample specific drug combination prediction model.
To start generalized prediction assignment, you should choose L1 in parameter model. Then drug that you are interested in predict combinable drugs should be input. Here, we provide two possible solutions to input the drug you are interested in DComboNet:
1) Input one or more drug name(s) together with their drugbank ID in a data.frame. In this case, you should shield manual input by manual_input = FALSE.
# runing level one model (L1)
drugcandidate = data.frame(drug = "Sorafenib", drugbankID = "DB00398")
DComboNet(load_dir = load_dir,
resultdir = resultdir,
model = "L1", # To choose level one model
manual_input = FALSE, # To shield manually input drug name
drugcandidate = drugcandidate,
drugnetWeight = TRUE, # Confirm if drug network should be weighted
featuretype = 'integrated_score') # Select which drug-drug similarity should be use
Note that if neither has drugcandidate been inputed nor has manual_input been allowed, all drug nodes in constructed network will be traversed for prediction.
2) Submit the name of drug in an interactive manner. Drug for prediction can be entered interactively after the question ‘Please type in the drug you are interested:’ pop out. Note that, if manual_input = TRUE, even with input drugcandidate, the final result will be generates for the manually input drug.
DComboNet(load_dir = load_dir,
resultdir = resultdir,
model = "L1",
manual_input = TRUE, # Select manually input function
drugcandidate = NULL,
drugnetWeight = TRUE,
featuretype = 'integrated_score')
# Don't run here
# Please type in the drug you are interested:
# Sorafenib
Three results will be generated and saved under the folder drugrank, generank and pathwayrank in resultdir path. Prediction result for drug candidates is saved in drugrank folder. The name of result file indicate drugseed and inside the file, drug candidates within network is ranked based on global similarity.
# Take Sorafenib as an example
model = "L1"
drugrank = read.csv(paste0(resultdir,model,"_result/drugrank/Sorafenib_rank.csv"))
DT::datatable(drugrank, options = list(pageLength = 5))Genes and pathways are also ranked according to global similarity. This two files will be used for subnetwork extraction and visualization.
# Take Sorafenib as an example
generank = read.csv(paste0(resultdir,model,"_result/generank/Sorafenib_rank.csv"))
DT::datatable(generank, options = list(pageLength = 5))
pathwayrank = read.csv(paste0(resultdir,model,"_result/pathwayrank/Sorafenib_rank.csv"))
DT::datatable(pathwayrank, options = list(pageLength = 5))
To run cancer sample specific model requires more input parameters, especially to define cancer cell line and drug treatment related variables. Drug induced gene expression changes are from LINCS database where two datasets (GSE70138 and GSE92742) are provided from GEO database. Multiple cancer cell lines together with different treatment time are also included in the two datasets. Depends on the specific purpose, you can choose cell line, treatment time and filtering criteria to generate your own gene set.
Still take Sorafenib as example, if aiming at predicting combinable drug for Sorafenib in Hepatocellular carcinoma cell line HEPG2, since corresponding dataset for HEPG2 is provided in LINCS and CCLE database, you can choose dataset, cellline and treatment_time for calling the drug-induced differential expressed gene table and HEPG2 specific expressed gene table, then the model can be set as follow:
# Prepare input drugseed
drugcandidate = data.frame(drug = "Sorafenib", drugbankID = "DB00398")
# The setting of other parameters can be the same as how we run level one model
DComboNet(load_dir = load_dir,
resultdir = resultdir,
model = "L2", # Choose level two model
manual_input = FALSE, # You can use manual input function (see level one model example)
drugcandidate = drugcandidate,
drugnetWeight = TRUE,
featuretype = 'integrated_score',
dataset = "92742",
cellline = "HEPG2",
treatment_time = 6,
# the absolute value of fold-change between drug-treated group and control group will be above 0.5
foldchange = 0.5,
# the p-value from the significent test (t.test) between drug-treated group and control group will be below 0.05
pvalue = 0.05)
If you have other gene list related to the cancer cell line you are interested in and want to included in drug-gene assocaition network and/or gene-gene association network, you can input drugDEG and cancergene table (in data.frame format) manually.
Similar to level one model, three results will be generated and saved under the folder drugrank, generank and pathwayrank in resultdir path. Prediction result for drug candidates is saved in drugrank folder. The name of result file indicate drugseed and inside the file, drug candidates within network is ranked based on global similarity.
# Take Sorafenib as an example
model = "L2"
drugrank = read.csv(paste0(resultdir,model,"_result/drugrank/Sorafenib_rank.csv"))
DT::datatable(drugrank, options = list(pageLength = 5))Genes and pathways are also ranked according to global similarity. This two files will be used for subnetwork extraction and visualization.
# Take Sorafenib as an example
generank = read.csv(paste0(resultdir,model,"_result/generank/Sorafenib_rank.csv"))
DT::datatable(generank, options = list(pageLength = 5))
pathwayrank = read.csv(paste0(resultdir,model,"_result/pathwayrank/Sorafenib_rank.csv"))
DT::datatable(pathwayrank, options = list(pageLength = 5))
Other than prediction function, DComboNet package provides functions to extract and visualize subnetwork between user-interested drug seed and its corresponding predicted combinable drug candidate, which may help infer the possible mechanism of drug combinations. This step only works when prediction assignment finished.
To run network visualization function, you need to check if the prediction result has been saved to given path resultdir. Note that the result files including drugrank table, generank table and pathwayrank table. Then capitalized drug names that you are interested can be inputed as drugseed and drugcandidate. drugtarget table should also be prepared as a dataframe and take as input. Genes and pathways for subnetwork construction are according to their rank corresponding to drugseed, the default value of gene rank (controlled via parametre generank_filter)is 0.01 meaning that only genes ranked on top 1% will be kept in subnetwork while default value of pathway rank(controlled via parametre pathwayrank_filter) is 0.1 meaning that only pathway ranked on top 10% will be kept in subnetwork.
After preparation of network visualization input, network_visualization() is used for visualizing network. This function was built based on visNetwork package. Drugseed, drugcandidate and their targeted genes are colorred coded. User can either click to select interesting nodes and drag around for better layout or select by id or group. An .html file will be generated and saved in under the result path. Furthermore, a .graphml file will also be saved. This file can be easily import as a network in Cytoscape and network style can be shown via choosing column type (e.g. color) and Mapping type set as “Passthrough Mapping”.
library(visNetwork)
drugseed = "Sorafenib"
drugcandidate = "Vorinostat"
drugtarget = data.frame(drug = c(rep(drugseed,10),rep(drugcandidate,5)),
target = c("BRAF", "FGFR1", "FLT1", "FLT3", "FLT4", "KDR", "KIT", "PDGFRB", "RAF1", "RET", "HDAC1", "HDAC2", "HDAC3", "HDAC6", "HDAC8"))
model = "L2"
network_extract(drugseed = drugseed,
drugcandidate = drugcandidate,
drugtarget = drugtarget,
generank_filter = 0.01,
pathwayrank_filter = 0.1,
model = model,
cellline = "HEPG2",
load_dir = load_dir,
resultdir = resultdir)
network_visualization(drugseed = drugseed,
drugcandidate = drugcandidate,
drugtarget = drugtarget,
model = model,
cellline = "HEPG2",
load_dir = load_dir,
resultdir = resultdir)Drug-drug pharmacological score table was provided for drug-drug association network:
library(DT)
drugnet_feature = read.table(paste0(load_dir,"drug_net/features.csv"), sep=",",header=TRUE, stringsAsFactors = FALSE)
# To fit the page, the values showing in the table below only kept two decimal places
# DT::datatable(drugnet_feature[1:20,], options = list(pageLength = 5))
# Only first 20 rows will be shown here as example.
The way to generate same feature table is packed in function DComboNet, user can also call function DrugNetFeature. To obtain features for newly inputed drug(s), extra data need to provide:
# druglist including drug names and their drugbank ID, note that the drug name better use name provided in drugbank
druglist = read.csv(paste0(load_dir, "/druglist_example.csv"), sep = ",", header = T, stringsAsFactors = F)
# DT::datatable(druglist, options = list(pageLength = 5))# For level one model
cellline = NULL
model = "L1"
# For level two model
# cellline = "HEPG2"
# model = "L2"
# feature will saved under the name of cell line
CDK_FP <- read.csv(paste0(load_dir,"data/fingerprints/fingerprints.csv"), sep = ",", header = T, stringsAsFactors = F)
pubchemFP <- read.csv(paste0(load_dir,"data/fingerprints/pubchem_fingerprints.csv"), sep = ",", header = T, stringsAsFactors = F)
MACCS_FP<- read.csv(paste0(load_dir,"data/fingerprints/MACCS_fingerprints.csv"), sep = ",", header = T, stringsAsFactors = F)
# Otherwise, please input PaDEL generated three fingerprint files
drugnet_feature = DComboNet::DrugNetFeature(druglist = druglist,
cellline = NULL,
model = 'L1',
CDK_FP=CDK_FP,
pubchemFP=pubchemFP,
MACCS_FP=MACCS_FP,
load_dir = load_dir)For level one model, drug-target gene associations and drug-related gene associations contribute to drug-gene association network:
drugtarget = read.table(paste0(load_dir,'data/drugtarget.csv'), sep =',', header = TRUE, stringsAsFactors = FALSE)
# DT::datatable(drugtarget[1:20,], options = list(pageLength = 5))
# Only first 20 rows will be shown here as example.druggene = unique(read.csv(paste0(load_dir,'data/drug_gene/drug_ipa.inPPI.csv'), header = T, stringsAsFactors = F)[c(1,3)])
# DT::datatable(druggene[1:20, ], options = list(pageLength = 5))
# Only first 20 rows will be shown here as example.For level two model, drug-related gene associations will be replaced by drug induced differentially expressed genes(DEG):
cellline = 'HEPG2'
dataset = '92742'
treatment_time = 6
drugDEG <- unique(read.csv(paste0(load_dir,'/LINCS_data/pathway_enrich_gsva_GSE',dataset,'/',cellline,'_DEG_',treatment_time,'h.csv'), header = T, stringsAsFactors = F))
# DT::datatable(drugDEG[1:20, ], options = list(pageLength = 5))
# Only first 20 rows will be shown here as example.
To generate drug-DEG association file, we provide function DEG_DEP_preparation() to obtain gene expression changes before and after by drug(s), then drugDEG_preparation() to filter with certain criteria.
druglist = read.csv(paste0(load_dir, "/druglist_example.csv"), sep = ",", header = T, stringsAsFactors = F)
cellline = 'HEPG2'
dataset = "92742" # which LINCS dataset to use
t = 6 # treatment time
# Before runing the function, please check how many available core can be use for calculation
# library(parallel)
# num_cores<-detectCores(logical=F)
num_cores = 2
# provide path to save results, please make sure it is saved in the same path with other modeling datasets
load_dir = "G:/lab/DCcomboNet/Rpackage/input_data/"
DEG_DEP_preparation(druglist = druglist,
dataset = dataset,
cellline = cellline,
treatment_time = t,
core = num_cores,
load_dir = load_dir)
drugDEG_preparation(cellline = cellline,
dataset = dataset,
treatment_time = t,
foldchange = 0.5,
pvalue = 0.05,
load_dir = load_dir)Gene-gene association network contains both cancer-related and drug-related genes. For level one model, cancers related genes were extracted from KEGG cancer related pathway including ‘pathway in cancer’:
cancer_gene = read.csv(paste0(load_dir,'data/network/protein-coding cancer genes name.csv'),header = F, stringsAsFactors = F)
head(cancer_gene[,1],10)
#> [1] "1433B" "1433E" "1433F" "1433G" "1433S" "1433T" "1433Z" "1A01" "1A02"
#> [10] "1A03"For level two model, cancer sample/cell line specific expressed gene list extracted from CCLE database is provided:
cellline = "HEPG2"
load_dir = "G:/lab/DCcomboNet/Rpackage/input_data/"
cancer_gene = read.csv(paste0(load_dir,'data/cellline_genes/',cellline,'_genelist.csv', sep = ''),header = T, stringsAsFactors = F)
head(cancer_gene[,1], 20)
#> [1] "AKT3" "NAALAD2" "NAALADL1" "CDH2" "TMEM170B"
#> [6] "LOC100127983" "TSTD1" "LOC100131607" "KIAA0664L3" "PEG3-AS1"
#> [11] "C17orf96" "CDH3" "LOC100287482" "LOC100287616" "LOC100287628"
#> [16] "LGALS8-AS1" "LOC100288911" "LOC100289092" "MCF2L-AS1" "MAMLD1"
If the gene list is not provided but the cancer cell line is included in CCLE database, function cancerGene() can be used to generate cancer cell line specific expressed gene list:
# provide path to save results, please make sure it is saved in the same path with other modeling datasets
cancer_gene = cancerGene(cellline = cellline,
load_dir = load_dir)
For level one model, the associations between drugs and pathways are via a target-pathway mapping solution, that is, the drug-pathway association exists if the target gene(s) of a drug in drug-drug association network can be mapped to pathway(s). This process is packed in function drugGeneNet.L1() when construct drug-gene association network adjacency matrix. If new drugs are inputed, in this step, drug-target genes table should be provided.
For level two model, pathways(DEpathway) that differentially regulated by drugs was added. Drug-DEpathway associations are obtained by gsva() function. The process can be done via function DEG_DEP_preparation() and then use certain filter criteria to select pathways through drugDEP_preparation():
druglist = read.csv(paste0(load_dir, "/druglist_example.csv"), sep = ",", header = T, stringsAsFactors = F)
cellline = 'HEPG2'
dataset = "92742" # which LINCS dataset to use
t = 6 # treatment time
# Before runing the function, please check how many available core can be use for calculation
# library(parallel)
# num_cores<-detectCores(logical=F)
num_cores = 2
# provide path to save results, please make sure it is saved in the same path with other modeling datasets
load_dir = "G:/lab/DCcomboNet/Rpackage/input_data/"
DEG_DEP_preparation(druglist = druglist,
dataset = dataset,
cellline = cellline,
treatment_time = t,
core = num_cores,
load_dir = load_dir)
drugDEP_preparation(cellline = cellline,
dataset = dataset,
treatment_time = t,
foldchange = 0.5,
pvalue = 0.05,
load_dir = load_dir)Some drug-DEpahtway tables are provided:
cellline = 'HEPG2'
dataset = '92742'
treatment_time = 6
drugDEP <- unique(read.csv(paste0(load_dir,'/LINCS_data/pathway_enrich_gsva_GSE',dataset,'/',cellline,'_DEgene_',treatment_time,'h.csv'), header = T, stringsAsFactors = F))
# DT::datatable(drugDEP[1:20, ], options = list(pageLength = 5))
# Only first 20 rows will be shown here as example.
After learning the pre-prepared network data files, let’s try some more prediction task. In session 4, we tried some basic usage of prediction function DComboNet with drug that is within our provided dataset. We provided some more ways to personalized models for your dataset.
The first kind is a simple extension for drug within provided dataset. Let’s continue taking “Sorafenib” as example. If there are new targeted genes or related gene got discovered (those genes that is not from Drugbank or IPA tools, or new interesting genes from publications), you can prepare your interested drug-targeted gene table (or drug-related gene table). For level one model, the prediction task can be as follow:
# runing level one model (L1)
drugcandidate = data.frame(drug = "Sorafenib", drugbankID = "DB00398")
# only as example
dt_table = data.frame(drug = 'Sorafenib', target = 'RAF1')
# only as example
dg_table = data.frame(drug = 'Sorafenib', gene = 'RAF1')
DComboNet(load_dir = load_dir,
resultdir = resultdir,
model = "L1", # To choose level one model
manual_input = FALSE, # To shield manually input drug name
drugcandidate = drugcandidate,
drugnetWeight = TRUE, # Confirm if drug network should be weighted
featuretype = 'integrated_score',
drugtarget = dt_table,
druggene = dg_table) # Select which drug-drug similarity should be use
Similarly, level two model can be run as:
# The setting of other parameters can be the same as how we run level one model
DComboNet(load_dir = load_dir,
resultdir = resultdir,
model = "L2", # Choose level two model
manual_input = FALSE, # You can use manual input function (see level one model example)
drugcandidate = drugcandidate,
drugnetWeight = TRUE,
featuretype = 'integrated_score',
drugtarget = dt_table,
druggene = dg_table,
dataset = "92742",
cellline = "HEPG2",
treatment_time = 6,
# the absolute value of fold-change between drug-treated group and control group will be above 0.5
foldchange = 0.5,
# the p-value from the significent test (t.test) between drug-treated group and control group will be below 0.05
pvalue = 0.05)
For level two model, if you have other processed drug-DEG table and drug-DEP for cell line ‘HEPG2’ and/or HEPG2 specific expressed gene list, you can also provided them in the model:
# only as example
drugDEG = read.table(paste0(load_dir, '/test/Sorafenib_DEG_test.txt'), sep = '\t', header = T, stringsAsFactors = F)
drugDEP = read.table(paste0(load_dir, '/test/Sorafenib_DEP_test.txt'), sep = '\t', header = T, stringsAsFactors = F)
cancergene = read.table(paste0(load_dir, '/test/HEPG2_genelist.csv'), sep = ',', header = T, stringsAsFactors = F)
# The setting of other parameters can be the same as how we run level one model
DComboNet(load_dir = load_dir,
resultdir = resultdir,
model = "L2", # Choose level two model
manual_input = FALSE, # You can use manual input function (see level one model example)
drugcandidate = drugcandidate,
drugnetWeight = TRUE,
featuretype = 'integrated_score',
drugtarget = dt_table,
druggene = dg_table,
cellline = "HEPG2", # You must provide cell line name for saving results and generating network
drugDEG = drugDEG,
drugDEP = drugDEP,
cancergene = cancergene)
DComboNet offers the possibility to extend network. You can provided your own drug list, drug-target data table, drug-related gene data table, drug-DEG/DEP tables etc to custmoized model to fit your purpose. In 5.2, you have learned how to provided extra drug-gene information for drug within provided dataset. Here, we are going to extend drug network. Take OCI-LY3 dataset as example.
drugpair_dt = read.csv(paste0(load_dir,'test/OCILY3_data/drugpair_oci_ly3.csv'),stringsAsFactors = F)
DT::datatable(drugpair_dt, options = list(pageLength = 3))drugcandidate <- read.csv(paste0(load_dir,"test/OCILY3_data/druglist_oci_ly3.csv"), sep = ',', header = T, stringsAsFactors = F)
DT::datatable(drugcandidate, options = list(pageLength = 3))CDK_FP <- read.csv(paste0(load_dir,'test/OCILY3_data/structure_sim/fingerprint.csv'), sep = ',',header = T, stringsAsFactors = F)
DT::datatable(CDK_FP[,1:5], options = list(pageLength = 3))pubchemFP <- read.csv(paste0(load_dir,'test/OCILY3_data/structure_sim/PubChem.csv'), sep = ',',header = T, stringsAsFactors = F)
DT::datatable(pubchemFP[,1:5], options = list(pageLength = 3))MACCS_FP <- read.csv(paste0(load_dir,'test/OCILY3_data/structure_sim/MACCF.csv'), sep = ',',header = T, stringsAsFactors = F)
DT::datatable(MACCS_FP[,1:5], options = list(pageLength = 3))drugtarget = read.csv(paste0(load_dir,'test/OCILY3_data/drug_target_oci_ly3.csv'), sep = ',',header = T, stringsAsFactors = F)
DT::datatable(drugtarget, options = list(pageLength = 3))drugDEG = read.table(paste0(load_dir,'test/OCILY3_data/drug_DEG_12hrs.txt'),sep='\t',header=T,stringsAsFactors = F)
DT::datatable(drugDEG, options = list(pageLength = 3))drugDEP = read.table(paste0(load_dir,'test/OCILY3_data/drug_DEP_12hrs.txt'),sep='\t',header=T,stringsAsFactors = F)
DT::datatable(drugDEP, options = list(pageLength = 3))cancergene = read.table(paste0(load_dir,'test/OCILY3_data/OCILY3_genelist.txt'),sep='\t',header=T,stringsAsFactors = F)
DT::datatable(cancergene, options = list(pageLength = 3))
DComboNet(load_dir = load_dir,
resultdir = resultdir,
model = 'L2',
manual_input = FALSE,
drugcandidate=drugcandidate,
CDK_FP = CDK_FP,
pubchemFP = pubchemFP,
MACCS_FP = MACCS_FP,
drugnetWeight = TRUE,
featuretype = "integrated_score",
drugtarget = drugtarget,
cellline = 'OCILY3',
drugDEG = drugDEG,
drugDEG = drugDEP,
cancergene = cancergene
)